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_exception"><tt>llvm.eh.exception</tt></a></li>
37 <li><a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a></li>
38 <li><a href="#llvm_eh_typeid_for"><tt>llvm.eh.typeid.for</tt></a></li>
39 <li><a href="#llvm_eh_sjlj_setjmp"><tt>llvm.eh.sjlj.setjmp</tt></a></li>
40 <li><a href="#llvm_eh_sjlj_longjmp"><tt>llvm.eh.sjlj.longjmp</tt></a></li>
41 <li><a href="#llvm_eh_sjlj_lsda"><tt>llvm.eh.sjlj.lsda</tt></a></li>
42 <li><a href="#llvm_eh_sjlj_callsite"><tt>llvm.eh.sjlj.callsite</tt></a></li>
43 <li><a href="#llvm_eh_sjlj_dispatchsetup"><tt>llvm.eh.sjlj.dispatchsetup</tt></a></li>
45 <li><a href="#asm">Asm Table Formats</a>
47 <li><a href="#unwind_tables">Exception Handling Frame</a></li>
48 <li><a href="#exception_tables">Exception Tables</a></li>
50 <li><a href="#todo">ToDo</a></li>
55 <div class="doc_author">
56 <p>Written by <a href="mailto:jlaskey@mac.com">Jim Laskey</a></p>
60 <!-- *********************************************************************** -->
61 <h2><a name="introduction">Introduction</a></h2>
62 <!-- *********************************************************************** -->
66 <p>This document is the central repository for all information pertaining to
67 exception handling in LLVM. It describes the format that LLVM exception
68 handling information takes, which is useful for those interested in creating
69 front-ends or dealing directly with the information. Further, this document
70 provides specific examples of what exception handling information is used for
73 <!-- ======================================================================= -->
75 <a name="itanium">Itanium ABI Zero-cost Exception Handling</a>
80 <p>Exception handling for most programming languages is designed to recover from
81 conditions that rarely occur during general use of an application. To that
82 end, exception handling should not interfere with the main flow of an
83 application's algorithm by performing checkpointing tasks, such as saving the
84 current pc or register state.</p>
86 <p>The Itanium ABI Exception Handling Specification defines a methodology for
87 providing outlying data in the form of exception tables without inlining
88 speculative exception handling code in the flow of an application's main
89 algorithm. Thus, the specification is said to add "zero-cost" to the normal
90 execution of an application.</p>
92 <p>A more complete description of the Itanium ABI exception handling runtime
93 support of can be found at
94 <a href="http://www.codesourcery.com/cxx-abi/abi-eh.html">Itanium C++ ABI:
95 Exception Handling</a>. A description of the exception frame format can be
97 <a href="http://refspecs.freestandards.org/LSB_3.0.0/LSB-Core-generic/LSB-Core-generic/ehframechpt.html">Exception
98 Frames</a>, with details of the DWARF 3 specification at
99 <a href="http://www.eagercon.com/dwarf/dwarf3std.htm">DWARF 3 Standard</a>.
100 A description for the C++ exception table formats can be found at
101 <a href="http://www.codesourcery.com/cxx-abi/exceptions.pdf">Exception Handling
106 <!-- ======================================================================= -->
108 <a name="sjlj">Setjmp/Longjmp Exception Handling</a>
113 <p>Setjmp/Longjmp (SJLJ) based exception handling uses LLVM intrinsics
114 <a href="#llvm_eh_sjlj_setjmp"><tt>llvm.eh.sjlj.setjmp</tt></a> and
115 <a href="#llvm_eh_sjlj_longjmp"><tt>llvm.eh.sjlj.longjmp</tt></a> to
116 handle control flow for exception handling.</p>
118 <p>For each function which does exception processing, be it try/catch blocks
119 or cleanups, that function registers itself on a global frame list. When
120 exceptions are being unwound, the runtime uses this list to identify which
121 functions need processing.<p>
123 <p>Landing pad selection is encoded in the call site entry of the function
124 context. The runtime returns to the function via
125 <a href="#llvm_eh_sjlj_longjmp"><tt>llvm.eh.sjlj.longjmp</tt></a>, where
126 a switch table transfers control to the appropriate landing pad based on
127 the index stored in the function context.</p>
129 <p>In contrast to DWARF exception handling, which encodes exception regions
130 and frame information in out-of-line tables, SJLJ exception handling
131 builds and removes the unwind frame context at runtime. This results in
132 faster exception handling at the expense of slower execution when no
133 exceptions are thrown. As exceptions are, by their nature, intended for
134 uncommon code paths, DWARF exception handling is generally preferred to
138 <!-- ======================================================================= -->
140 <a name="overview">Overview</a>
145 <p>When an exception is thrown in LLVM code, the runtime does its best to find a
146 handler suited to processing the circumstance.</p>
148 <p>The runtime first attempts to find an <i>exception frame</i> corresponding to
149 the function where the exception was thrown. If the programming language
150 (e.g. C++) supports exception handling, the exception frame contains a
151 reference to an exception table describing how to process the exception. If
152 the language (e.g. C) does not support exception handling, or if the
153 exception needs to be forwarded to a prior activation, the exception frame
154 contains information about how to unwind the current activation and restore
155 the state of the prior activation. This process is repeated until the
156 exception is handled. If the exception is not handled and no activations
157 remain, then the application is terminated with an appropriate error
160 <p>Because different programming languages have different behaviors when
161 handling exceptions, the exception handling ABI provides a mechanism for
162 supplying <i>personalities.</i> An exception handling personality is defined
163 by way of a <i>personality function</i> (e.g. <tt>__gxx_personality_v0</tt>
164 in C++), which receives the context of the exception, an <i>exception
165 structure</i> containing the exception object type and value, and a reference
166 to the exception table for the current function. The personality function
167 for the current compile unit is specified in a <i>common exception
170 <p>The organization of an exception table is language dependent. For C++, an
171 exception table is organized as a series of code ranges defining what to do
172 if an exception occurs in that range. Typically, the information associated
173 with a range defines which types of exception objects (using C++ <i>type
174 info</i>) that are handled in that range, and an associated action that
175 should take place. Actions typically pass control to a <i>landing
178 <p>A landing pad corresponds to the code found in the <i>catch</i> portion of
179 a <i>try</i>/<i>catch</i> sequence. When execution resumes at a landing
180 pad, it receives the exception structure and a selector corresponding to
181 the <i>type</i> of exception thrown. The selector is then used to determine
182 which <i>catch</i> should actually process the exception.</p>
188 <!-- ======================================================================= -->
190 <a name="codegen">LLVM Code Generation</a>
195 <p>At the time of this writing, only C++ exception handling support is available
196 in LLVM. So the remainder of this document will be somewhat C++-centric.</p>
198 <p>From the C++ developers perspective, exceptions are defined in terms of the
199 <tt>throw</tt> and <tt>try</tt>/<tt>catch</tt> statements. In this section
200 we will describe the implementation of LLVM exception handling in terms of
203 <!-- ======================================================================= -->
205 <a name="throw">Throw</a>
210 <p>Languages that support exception handling typically provide a <tt>throw</tt>
211 operation to initiate the exception process. Internally, a throw operation
212 breaks down into two steps. First, a request is made to allocate exception
213 space for an exception structure. This structure needs to survive beyond the
214 current activation. This structure will contain the type and value of the
215 object being thrown. Second, a call is made to the runtime to raise the
216 exception, passing the exception structure as an argument.</p>
218 <p>In C++, the allocation of the exception structure is done by
219 the <tt>__cxa_allocate_exception</tt> runtime function. The exception
220 raising is handled by <tt>__cxa_throw</tt>. The type of the exception is
221 represented 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: where to continue when the call succeeds
236 as per normal; and where to continue if the call raises an exception, either
237 by a throw or the unwinding of a throw.</p>
239 <p>The term used to define a the place where an <tt>invoke</tt> continues after
240 an exception is called a <i>landing pad</i>. LLVM landing pads are
241 conceptually alternative function entry points where an exception structure
242 reference and a type info index are passed in as arguments. The landing pad
243 saves the exception structure reference and then proceeds to select the catch
244 block that corresponds to the type info of the exception object.</p>
246 <p>Two LLVM intrinsic functions are used to convey information about the landing
247 pad to the back end.</p>
250 <li><a href="#llvm_eh_exception"><tt>llvm.eh.exception</tt></a> takes no
251 arguments and returns a pointer to the exception structure. This only
252 returns a sensible value if called after an <tt>invoke</tt> has branched
253 to a landing pad. Due to code generation limitations, it must currently
254 be called in the landing pad itself.</li>
256 <li><a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> takes a minimum
257 of three arguments. The first argument is the reference to the exception
258 structure. The second argument is a reference to the personality function
259 to be used for this <tt>try</tt>/<tt>catch</tt> sequence. Each of the
260 remaining arguments is either a reference to the type info for
261 a <tt>catch</tt> statement, a <a href="#throw_filters">filter</a>
262 expression, or the number zero (<tt>0</tt>) representing
263 a <a href="#cleanups">cleanup</a>. The exception is tested against the
264 arguments sequentially from first to last. The result of
265 the <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> is a
266 positive number if the exception matched a type info, a negative number if
267 it matched a filter, and zero if it matched a cleanup. If nothing is
268 matched, the behaviour of the program
269 is <a href="#restrictions">undefined</a>. This only returns a sensible
270 value if called after an <tt>invoke</tt> has branched to a landing pad.
271 Due to codegen limitations, it must currently be called in the landing pad
272 itself. If a type info matched, then the selector value is the index of
273 the type info in the exception table, which can be obtained using the
274 <a href="#llvm_eh_typeid_for"><tt>llvm.eh.typeid.for</tt></a>
278 <p>Once the landing pad has the type info selector, the code branches to the
279 code for the first catch. The catch then checks the value of the type info
280 selector against the index of type info for that catch. Since the type info
281 index is not known until all the type info have been gathered in the backend,
282 the catch code will call the
283 <a href="#llvm_eh_typeid_for"><tt>llvm.eh.typeid.for</tt></a> intrinsic
284 to determine the index for a given type info. If the catch fails to match
285 the selector then control is passed on to the next catch. Note: Since the
286 landing pad will not be used if there is no match in the list of type info on
287 the call to <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a>, then
288 neither the last catch nor <i>catch all</i> need to perform the check
289 against the selector.</p>
291 <p>Finally, the entry and exit of catch code is bracketed with calls
292 to <tt>__cxa_begin_catch</tt> and <tt>__cxa_end_catch</tt>.</p>
295 <li><tt>__cxa_begin_catch</tt> takes a exception structure reference as an
296 argument and returns the value of the exception object.</li>
298 <li><tt>__cxa_end_catch</tt> takes no arguments. This function:<br><br>
300 <li>Locates the most recently caught exception and decrements its handler
302 <li>Removes the exception from the "caught" stack if the handler count
303 goes to zero, and</li>
304 <li>Destroys the exception if the handler count goes to zero, and the
305 exception was not re-thrown by throw.</li>
307 <p>Note: a rethrow from within the catch may replace this call with
308 a <tt>__cxa_rethrow</tt>.</p></li>
313 <!-- ======================================================================= -->
315 <a name="cleanups">Cleanups</a>
320 <p>To handle destructors and cleanups in <tt>try</tt> code, control may not run
321 directly from a landing pad to the first catch. Control may actually flow
322 from the landing pad to clean up code and then to the first catch. Since the
323 required clean up for each <tt>invoke</tt> in a <tt>try</tt> may be different
324 (e.g. intervening constructor), there may be several landing pads for a given
325 try. If cleanups need to be run, an <tt>i32 0</tt> should be passed as the
326 last <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> argument.
327 However, when using DWARF exception handling with C++, a <tt>i8* null</tt>
328 <a href="#restrictions">must</a> be passed instead.</p>
332 <!-- ======================================================================= -->
334 <a name="throw_filters">Throw Filters</a>
339 <p>C++ allows the specification of which exception types can be thrown from a
340 function. To represent this a top level landing pad may exist to filter out
341 invalid types. To express this in LLVM code the landing pad will
342 call <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a>. The
343 arguments are a reference to the exception structure, a reference to the
344 personality function, the length of the filter expression (the number of type
345 infos plus one), followed by the type infos themselves.
346 <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> will return a
347 negative value if the exception does not match any of the type infos. If no
348 match is found then a call to <tt>__cxa_call_unexpected</tt> should be made,
349 otherwise <tt>_Unwind_Resume</tt>. Each of these functions requires a
350 reference to the exception structure. Note that the most general form of an
351 <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> call can contain
352 any number of type infos, filter expressions and cleanups (though having more
353 than one cleanup is pointless). The LLVM C++ front-end can generate such
354 <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> calls due to
355 inlining creating nested exception handling scopes.</p>
359 <!-- ======================================================================= -->
361 <a name="restrictions">Restrictions</a>
366 <p>The semantics of the invoke instruction require that any exception that
367 unwinds through an invoke call should result in a branch to the invoke's
368 unwind label. However such a branch will only happen if the
369 <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> matches. Thus in
370 order to ensure correct operation, the front-end must only generate
371 <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> calls that are
372 guaranteed to always match whatever exception unwinds through the invoke.
373 For most languages it is enough to pass zero, indicating the presence of
374 a <a href="#cleanups">cleanup</a>, as the
375 last <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> argument.
376 However for C++ this is not sufficient, because the C++ personality function
377 will terminate the program if it detects that unwinding the exception only
378 results in matches with cleanups. For C++ a <tt>null i8*</tt> should be
379 passed as the last <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a>
380 argument instead. This is interpreted as a catch-all by the C++ personality
381 function, and will always match.</p>
387 <!-- ======================================================================= -->
389 <a name="format_common_intrinsics">Exception Handling Intrinsics</a>
394 <p>LLVM uses several intrinsic functions (name prefixed with "llvm.eh") to
395 provide exception handling information at various points in generated
398 <!-- ======================================================================= -->
400 <a name="llvm_eh_exception">llvm.eh.exception</a>
406 i8* %<a href="#llvm_eh_exception">llvm.eh.exception</a>()
409 <p>This intrinsic returns a pointer to the exception structure.</p>
413 <!-- ======================================================================= -->
415 <a name="llvm_eh_selector">llvm.eh.selector</a>
421 i32 %<a href="#llvm_eh_selector">llvm.eh.selector</a>(i8*, i8*, ...)
424 <p>This intrinsic is used to compare the exception with the given type infos,
425 filters and cleanups.</p>
427 <p><a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> takes a minimum of
428 three arguments. The first argument is the reference to the exception
429 structure. The second argument is a reference to the personality function to
430 be used for this try catch sequence. Each of the remaining arguments is
431 either a reference to the type info for a catch statement,
432 a <a href="#throw_filters">filter</a> expression, or the number zero
433 representing a <a href="#cleanups">cleanup</a>. The exception is tested
434 against the arguments sequentially from first to last. The result of
435 the <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a> is a positive
436 number if the exception matched a type info, a negative number if it matched
437 a filter, and zero if it matched a cleanup. If nothing is matched, the
438 behaviour of the program is <a href="#restrictions">undefined</a>. If a type
439 info matched then the selector value is the index of the type info in the
440 exception table, which can be obtained using the
441 <a href="#llvm_eh_typeid_for"><tt>llvm.eh.typeid.for</tt></a> intrinsic.</p>
445 <!-- ======================================================================= -->
447 <a name="llvm_eh_typeid_for">llvm.eh.typeid.for</a>
453 i32 %<a href="#llvm_eh_typeid_for">llvm.eh.typeid.for</a>(i8*)
456 <p>This intrinsic returns the type info index in the exception table of the
457 current function. This value can be used to compare against the result
458 of <a href="#llvm_eh_selector"><tt>llvm.eh.selector</tt></a>. The single
459 argument is a reference to a type info.</p>
463 <!-- ======================================================================= -->
465 <a name="llvm_eh_sjlj_setjmp">llvm.eh.sjlj.setjmp</a>
471 i32 %<a href="#llvm_eh_sjlj_setjmp">llvm.eh.sjlj.setjmp</a>(i8*)
474 <p>The SJLJ exception handling uses this intrinsic to force register saving for
475 the current function and to store the address of the following instruction
476 for use as a destination address by <a href="#llvm_eh_sjlj_longjmp">
477 <tt>llvm.eh.sjlj.longjmp</tt></a>. The buffer format and the overall
478 functioning of this intrinsic is compatible with the GCC
479 <tt>__builtin_setjmp</tt> implementation, allowing code built with the
480 two compilers to interoperate.</p>
482 <p>The single parameter is a pointer to a five word buffer in which the calling
483 context is saved. The front end places the frame pointer in the first word,
484 and the target implementation of this intrinsic should place the destination
486 <a href="#llvm_eh_sjlj_longjmp"><tt>llvm.eh.sjlj.longjmp</tt></a> in the
487 second word. The following three words are available for use in a
488 target-specific manner.</p>
492 <!-- ======================================================================= -->
494 <a name="llvm_eh_sjlj_longjmp">llvm.eh.sjlj.longjmp</a>
500 void %<a href="#llvm_eh_sjlj_longjmp">llvm.eh.sjlj.setjmp</a>(i8*)
503 <p>The <a href="#llvm_eh_sjlj_longjmp"><tt>llvm.eh.sjlj.longjmp</tt></a>
504 intrinsic is used to implement <tt>__builtin_longjmp()</tt> for SJLJ
505 style exception handling. The single parameter is a pointer to a
506 buffer populated by <a href="#llvm_eh_sjlj_setjmp">
507 <tt>llvm.eh.sjlj.setjmp</tt></a>. The frame pointer and stack pointer
508 are restored from the buffer, then control is transferred to the
509 destination address.</p>
512 <!-- ======================================================================= -->
514 <a name="llvm_eh_sjlj_lsda">llvm.eh.sjlj.lsda</a>
520 i8* %<a href="#llvm_eh_sjlj_lsda">llvm.eh.sjlj.lsda</a>()
523 <p>Used for SJLJ based exception handling, the <a href="#llvm_eh_sjlj_lsda">
524 <tt>llvm.eh.sjlj.lsda</tt></a> intrinsic returns the address of the Language
525 Specific Data Area (LSDA) for the current function. The SJLJ front-end code
526 stores this address in the exception handling function context for use by the
531 <!-- ======================================================================= -->
533 <a name="llvm_eh_sjlj_callsite">llvm.eh.sjlj.callsite</a>
539 void %<a href="#llvm_eh_sjlj_callsite">llvm.eh.sjlj.callsite</a>(i32)
542 <p>For SJLJ based exception handling, the <a href="#llvm_eh_sjlj_callsite">
543 <tt>llvm.eh.sjlj.callsite</tt></a> intrinsic identifies the callsite value
544 associated with the following invoke instruction. This is used to ensure
545 that landing pad entries in the LSDA are generated in the matching order.</p>
549 <!-- ======================================================================= -->
551 <a name="llvm_eh_sjlj_dispatchsetup">llvm.eh.sjlj.dispatchsetup</a>
557 void %<a href="#llvm_eh_sjlj_dispatchsetup">llvm.eh.sjlj.dispatchsetup</a>(i32)
560 <p>For SJLJ based exception handling, the <a href="#llvm_eh_sjlj_dispatchsetup">
561 <tt>llvm.eh.sjlj.dispatchsetup</tt></a> intrinsic is used by targets to do
562 any unwind-edge setup they need. By default, no action is taken. </p>
568 <!-- ======================================================================= -->
570 <a name="asm">Asm Table Formats</a>
575 <p>There are two tables that are used by the exception handling runtime to
576 determine which actions should take place when an exception is thrown.</p>
578 <!-- ======================================================================= -->
580 <a name="unwind_tables">Exception Handling Frame</a>
585 <p>An exception handling frame <tt>eh_frame</tt> is very similar to the unwind
586 frame used by dwarf debug info. The frame contains all the information
587 necessary to tear down the current frame and restore the state of the prior
588 frame. There is an exception handling frame for each function in a compile
589 unit, plus a common exception handling frame that defines information common
590 to all functions in the unit.</p>
592 <p>Todo - Table details here.</p>
596 <!-- ======================================================================= -->
598 <a name="exception_tables">Exception Tables</a>
603 <p>An exception table contains information about what actions to take when an
604 exception is thrown in a particular part of a function's code. There is one
605 exception table per function except leaf routines and functions that have
606 only calls to non-throwing functions will not need an exception table.</p>
608 <p>Todo - Table details here.</p>
614 <!-- ======================================================================= -->
616 <a name="todo">ToDo</a>
623 <li>Testing/Testing/Testing.</li>
629 <!-- *********************************************************************** -->
633 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
634 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
635 <a href="http://validator.w3.org/check/referer"><img
636 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
638 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
639 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
640 Last modified: $Date$