1 .. _exception_handling:
3 ==========================
4 Exception Handling in LLVM
5 ==========================
13 This document is the central repository for all information pertaining to
14 exception handling in LLVM. It describes the format that LLVM exception
15 handling information takes, which is useful for those interested in creating
16 front-ends or dealing directly with the information. Further, this document
17 provides specific examples of what exception handling information is used for in
20 Itanium ABI Zero-cost Exception Handling
21 ----------------------------------------
23 Exception handling for most programming languages is designed to recover from
24 conditions that rarely occur during general use of an application. To that end,
25 exception handling should not interfere with the main flow of an application's
26 algorithm by performing checkpointing tasks, such as saving the current pc or
29 The Itanium ABI Exception Handling Specification defines a methodology for
30 providing outlying data in the form of exception tables without inlining
31 speculative exception handling code in the flow of an application's main
32 algorithm. Thus, the specification is said to add "zero-cost" to the normal
33 execution of an application.
35 A more complete description of the Itanium ABI exception handling runtime
36 support of can be found at `Itanium C++ ABI: Exception Handling
37 <http://www.codesourcery.com/cxx-abi/abi-eh.html>`_. A description of the
38 exception frame format can be found at `Exception Frames
39 <http://refspecs.freestandards.org/LSB_3.0.0/LSB-Core-generic/LSB-Core-generic/ehframechpt.html>`_,
40 with details of the DWARF 4 specification at `DWARF 4 Standard
41 <http://dwarfstd.org/Dwarf4Std.php>`_. A description for the C++ exception
42 table formats can be found at `Exception Handling Tables
43 <http://www.codesourcery.com/cxx-abi/exceptions.pdf>`_.
45 Setjmp/Longjmp Exception Handling
46 ---------------------------------
48 Setjmp/Longjmp (SJLJ) based exception handling uses LLVM intrinsics
49 `llvm.eh.sjlj.setjmp`_ and `llvm.eh.sjlj.longjmp`_ to handle control flow for
52 For each function which does exception processing --- be it ``try``/``catch``
53 blocks or cleanups --- that function registers itself on a global frame
54 list. When exceptions are unwinding, the runtime uses this list to identify
55 which functions need processing.
57 Landing pad selection is encoded in the call site entry of the function
58 context. The runtime returns to the function via `llvm.eh.sjlj.longjmp`_, where
59 a switch table transfers control to the appropriate landing pad based on the
60 index stored in the function context.
62 In contrast to DWARF exception handling, which encodes exception regions and
63 frame information in out-of-line tables, SJLJ exception handling builds and
64 removes the unwind frame context at runtime. This results in faster exception
65 handling at the expense of slower execution when no exceptions are thrown. As
66 exceptions are, by their nature, intended for uncommon code paths, DWARF
67 exception handling is generally preferred to SJLJ.
72 When an exception is thrown in LLVM code, the runtime does its best to find a
73 handler suited to processing the circumstance.
75 The runtime first attempts to find an *exception frame* corresponding to the
76 function where the exception was thrown. If the programming language supports
77 exception handling (e.g. C++), the exception frame contains a reference to an
78 exception table describing how to process the exception. If the language does
79 not support exception handling (e.g. C), or if the exception needs to be
80 forwarded to a prior activation, the exception frame contains information about
81 how to unwind the current activation and restore the state of the prior
82 activation. This process is repeated until the exception is handled. If the
83 exception is not handled and no activations remain, then the application is
84 terminated with an appropriate error message.
86 Because different programming languages have different behaviors when handling
87 exceptions, the exception handling ABI provides a mechanism for
88 supplying *personalities*. An exception handling personality is defined by
89 way of a *personality function* (e.g. ``__gxx_personality_v0`` in C++),
90 which receives the context of the exception, an *exception structure*
91 containing the exception object type and value, and a reference to the exception
92 table for the current function. The personality function for the current
93 compile unit is specified in a *common exception frame*.
95 The organization of an exception table is language dependent. For C++, an
96 exception table is organized as a series of code ranges defining what to do if
97 an exception occurs in that range. Typically, the information associated with a
98 range defines which types of exception objects (using C++ *type info*) that are
99 handled in that range, and an associated action that should take place. Actions
100 typically pass control to a *landing pad*.
102 A landing pad corresponds roughly to the code found in the ``catch`` portion of
103 a ``try``/``catch`` sequence. When execution resumes at a landing pad, it
104 receives an *exception structure* and a *selector value* corresponding to the
105 *type* of exception thrown. The selector is then used to determine which *catch*
106 should actually process the exception.
111 From a C++ developer's perspective, exceptions are defined in terms of the
112 ``throw`` and ``try``/``catch`` statements. In this section we will describe the
113 implementation of LLVM exception handling in terms of C++ examples.
118 Languages that support exception handling typically provide a ``throw``
119 operation to initiate the exception process. Internally, a ``throw`` operation
120 breaks down into two steps.
122 #. A request is made to allocate exception space for an exception structure.
123 This structure needs to survive beyond the current activation. This structure
124 will contain the type and value of the object being thrown.
126 #. A call is made to the runtime to raise the exception, passing the exception
127 structure as an argument.
129 In C++, the allocation of the exception structure is done by the
130 ``__cxa_allocate_exception`` runtime function. The exception raising is handled
131 by ``__cxa_throw``. The type of the exception is represented using a C++ RTTI
137 A call within the scope of a *try* statement can potentially raise an
138 exception. In those circumstances, the LLVM C++ front-end replaces the call with
139 an ``invoke`` instruction. Unlike a call, the ``invoke`` has two potential
142 #. where to continue when the call succeeds as per normal, and
144 #. where to continue if the call raises an exception, either by a throw or the
147 The term used to define a the place where an ``invoke`` continues after an
148 exception is called a *landing pad*. LLVM landing pads are conceptually
149 alternative function entry points where an exception structure reference and a
150 type info index are passed in as arguments. The landing pad saves the exception
151 structure reference and then proceeds to select the catch block that corresponds
152 to the type info of the exception object.
154 The LLVM `landingpad instruction <LangRef.html#i_landingpad>`_ is used to convey
155 information about the landing pad to the back end. For C++, the ``landingpad``
156 instruction returns a pointer and integer pair corresponding to the pointer to
157 the *exception structure* and the *selector value* respectively.
159 The ``landingpad`` instruction takes a reference to the personality function to
160 be used for this ``try``/``catch`` sequence. The remainder of the instruction is
161 a list of *cleanup*, *catch*, and *filter* clauses. The exception is tested
162 against the clauses sequentially from first to last. The selector value is a
163 positive number if the exception matched a type info, a negative number if it
164 matched a filter, and zero if it matched a cleanup. If nothing is matched, the
165 behavior of the program is `undefined`_. If a type info matched, then the
166 selector value is the index of the type info in the exception table, which can
167 be obtained using the `llvm.eh.typeid.for`_ intrinsic.
169 Once the landing pad has the type info selector, the code branches to the code
170 for the first catch. The catch then checks the value of the type info selector
171 against the index of type info for that catch. Since the type info index is not
172 known until all the type infos have been gathered in the backend, the catch code
173 must call the `llvm.eh.typeid.for`_ intrinsic to determine the index for a given
174 type info. If the catch fails to match the selector then control is passed on to
177 Finally, the entry and exit of catch code is bracketed with calls to
178 ``__cxa_begin_catch`` and ``__cxa_end_catch``.
180 * ``__cxa_begin_catch`` takes an exception structure reference as an argument
181 and returns the value of the exception object.
183 * ``__cxa_end_catch`` takes no arguments. This function:
185 #. Locates the most recently caught exception and decrements its handler
188 #. Removes the exception from the *caught* stack if the handler count goes to
191 #. Destroys the exception if the handler count goes to zero and the exception
192 was not re-thrown by throw.
196 a rethrow from within the catch may replace this call with a
202 A cleanup is extra code which needs to be run as part of unwinding a scope. C++
203 destructors are a typical example, but other languages and language extensions
204 provide a variety of different kinds of cleanups. In general, a landing pad may
205 need to run arbitrary amounts of cleanup code before actually entering a catch
206 block. To indicate the presence of cleanups, a `landingpad
207 instruction <LangRef.html#i_landingpad>`_ should have a *cleanup*
208 clause. Otherwise, the unwinder will not stop at the landing pad if there are no
209 catches or filters that require it to.
213 Do not allow a new exception to propagate out of the execution of a
214 cleanup. This can corrupt the internal state of the unwinder. Different
215 languages describe different high-level semantics for these situations: for
216 example, C++ requires that the process be terminated, whereas Ada cancels both
217 exceptions and throws a third.
219 When all cleanups are finished, if the exception is not handled by the current
220 function, resume unwinding by calling the `resume
221 instruction <LangRef.html#i_resume>`_, passing in the result of the
222 ``landingpad`` instruction for the original landing pad.
227 C++ allows the specification of which exception types may be thrown from a
228 function. To represent this, a top level landing pad may exist to filter out
229 invalid types. To express this in LLVM code the `landingpad
230 instruction <LangRef.html#i_landingpad>`_ will have a filter clause. The clause
231 consists of an array of type infos. ``landingpad`` will return a negative value
232 if the exception does not match any of the type infos. If no match is found then
233 a call to ``__cxa_call_unexpected`` should be made, otherwise
234 ``_Unwind_Resume``. Each of these functions requires a reference to the
235 exception structure. Note that the most general form of a ``landingpad``
236 instruction can have any number of catch, cleanup, and filter clauses (though
237 having more than one cleanup is pointless). The LLVM C++ front-end can generate
238 such ``landingpad`` instructions due to inlining creating nested exception
246 The unwinder delegates the decision of whether to stop in a call frame to that
247 call frame's language-specific personality function. Not all unwinders guarantee
248 that they will stop to perform cleanups. For example, the GNU C++ unwinder
249 doesn't do so unless the exception is actually caught somewhere further up the
252 In order for inlining to behave correctly, landing pads must be prepared to
253 handle selector results that they did not originally advertise. Suppose that a
254 function catches exceptions of type ``A``, and it's inlined into a function that
255 catches exceptions of type ``B``. The inliner will update the ``landingpad``
256 instruction for the inlined landing pad to include the fact that ``B`` is also
257 caught. If that landing pad assumes that it will only be entered to catch an
258 ``A``, it's in for a rude awakening. Consequently, landing pads must test for
259 the selector results they understand and then resume exception propagation with
260 the `resume instruction <LangRef.html#i_resume>`_ if none of the conditions
263 Exception Handling Intrinsics
264 =============================
266 In addition to the ``landingpad`` and ``resume`` instructions, LLVM uses several
267 intrinsic functions (name prefixed with ``llvm.eh``) to provide exception
268 handling information at various points in generated code.
270 .. _llvm.eh.typeid.for:
277 i32 @llvm.eh.typeid.for(i8* %type_info)
280 This intrinsic returns the type info index in the exception table of the current
281 function. This value can be used to compare against the result of
282 ``landingpad`` instruction. The single argument is a reference to a type info.
284 .. _llvm.eh.sjlj.setjmp:
291 i32 @llvm.eh.sjlj.setjmp(i8* %setjmp_buf)
293 For SJLJ based exception handling, this intrinsic forces register saving for the
294 current function and stores the address of the following instruction for use as
295 a destination address by `llvm.eh.sjlj.longjmp`_. The buffer format and the
296 overall functioning of this intrinsic is compatible with the GCC
297 ``__builtin_setjmp`` implementation allowing code built with the clang and GCC
300 The single parameter is a pointer to a five word buffer in which the calling
301 context is saved. The front end places the frame pointer in the first word, and
302 the target implementation of this intrinsic should place the destination address
303 for a `llvm.eh.sjlj.longjmp`_ in the second word. The following three words are
304 available for use in a target-specific manner.
306 .. _llvm.eh.sjlj.longjmp:
313 void @llvm.eh.sjlj.longjmp(i8* %setjmp_buf)
315 For SJLJ based exception handling, the ``llvm.eh.sjlj.longjmp`` intrinsic is
316 used to implement ``__builtin_longjmp()``. The single parameter is a pointer to
317 a buffer populated by `llvm.eh.sjlj.setjmp`_. The frame pointer and stack
318 pointer are restored from the buffer, then control is transferred to the
326 i8* @llvm.eh.sjlj.lsda()
328 For SJLJ based exception handling, the ``llvm.eh.sjlj.lsda`` intrinsic returns
329 the address of the Language Specific Data Area (LSDA) for the current
330 function. The SJLJ front-end code stores this address in the exception handling
331 function context for use by the runtime.
333 llvm.eh.sjlj.callsite
334 ---------------------
338 void @llvm.eh.sjlj.callsite(i32 %call_site_num)
340 For SJLJ based exception handling, the ``llvm.eh.sjlj.callsite`` intrinsic
341 identifies the callsite value associated with the following ``invoke``
342 instruction. This is used to ensure that landing pad entries in the LSDA are
343 generated in matching order.
348 There are two tables that are used by the exception handling runtime to
349 determine which actions should be taken when an exception is thrown.
351 Exception Handling Frame
352 ------------------------
354 An exception handling frame ``eh_frame`` is very similar to the unwind frame
355 used by DWARF debug info. The frame contains all the information necessary to
356 tear down the current frame and restore the state of the prior frame. There is
357 an exception handling frame for each function in a compile unit, plus a common
358 exception handling frame that defines information common to all functions in the
364 An exception table contains information about what actions to take when an
365 exception is thrown in a particular part of a function's code. There is one
366 exception table per function, except leaf functions and functions that have
367 calls only to non-throwing functions. They do not need an exception table.