1 CDSChecker: A Model Checker for C11 and C++11 Atomics
2 =====================================================
4 Copyright © 2013 Regents of the University of California. All rights reserved.
6 CDSChecker is distributed under the GPL v2. See the LICENSE file for details.
12 CDSChecker is a model checker for C11/C++11 which exhaustively explores the
13 behaviors of code under the C/C++ memory model. It uses partial order reduction
14 as well as a few other novel techniques to eliminate time spent on redundant
15 execution behaviors and to significantly shrink the state space. The model
16 checking algorithm is described in more detail in this paper (published in
19 > <http://demsky.eecs.uci.edu/publications/c11modelcheck.pdf>
21 It is designed to support unit tests on concurrent data structure written using
24 CDSChecker is constructed as a dynamically-linked shared library which
25 implements the C and C++ atomic types and portions of the other thread-support
26 libraries of C/C++ (e.g., std::atomic, std::mutex, etc.). Notably, we only
27 support the C version of threads (i.e., `thrd_t` and similar, from `<threads.h>`),
28 because C++ threads require features which are only available to a C++11
29 compiler (and we want to support others, at least for now).
31 CDSChecker should compile on Linux and Mac OSX with no dependencies and has been
32 tested with LLVM (clang/clang++) and GCC. It likely can be ported to other \*NIX
33 flavors. We have not attempted to port to Windows.
39 If you haven't done so already, you may download CDSChecker using
40 [git](http://git-scm.com/) (for those without git, snapshots can be found at the
43 git clone git://demsky.eecs.uci.edu/model-checker.git
45 Get the benchmarks (not required; distributed separately):
48 git clone git://demsky.eecs.uci.edu/model-checker-benchmarks.git benchmarks
50 Compile the model checker:
54 Compile the benchmarks:
58 Run a simple example (the `run.sh` script does some very minimal processing for
61 ./run.sh test/userprog.o
63 To see the help message on how to run CDSChecker, execute:
73 > Controls the liveness of the memory system. Note that multithreaded programs
74 > often rely on memory liveness for termination, so this parameter is
75 > necessary for such programs.
77 > Liveness is controlled by `num`: the number of times a load is allowed to
78 > see the same store when a newer store exists---one that is ordered later in
79 > the modification order.
83 > Turns on CHESS-like yield-based fairness support (requires `thrd_yield()`
84 > instrumentation in test program).
88 > Turns on alternative fairness support (less desirable than `-y`). A
89 > necessary alternative for some programs that do not support yield-based
94 > Verbose: show all executions and not just buggy ones.
98 > Constrain how long we will run to wait for a future value past when it is
103 > Value to provide to atomics loads from uninitialized memory locations. The
104 > default is 0, but this may cause some programs to throw exceptions
105 > (segfault) before the model checker prints a trace.
119 Many simple tests are located in the `tests/` directory. You may also want to
120 try the larger benchmarks (distributed separately), which can be placed under
121 the `benchmarks/` directory. After building CDSChecker, you can build and run
122 the benchmarks as follows:
127 > # run barrier test with fairness/memory liveness
128 > ./run.sh barrier/barrier -y -m 2
130 > # Linux reader/write lock test with fairness/memory liveness
131 > ./run.sh linuxrwlocks/linuxrwlocks -y -m 2
133 > # run all benchmarks and provide timing results
137 Running your own code
138 ---------------------
140 You likely want to test your own code, not just our simple tests. To do so, you
141 need to perform a few steps.
143 First, because CDSChecker executes your program dozens (if not hundreds or
144 thousands) of times, you will have the most success if your code is written as a
145 unit test and not as a full-blown program.
147 Second, because CDSChecker must be able to manage your program for you, your
148 program should declare its main entry point as `user_main(int, char**)` rather
149 than `main(int, char**)`.
151 Third, test programs must use the standard C11/C++11 library headers (see below
152 for supported APIs) and must compile against the versions provided in
153 CDSChecker's `include/` directory. Notably, we only support C11 thread syntax
154 (`thrd_t`, etc. from `<thread.h>`).
156 Test programs may also use our included happens-before race detector by
157 including <librace.h> and utilizing the appropriate functions
158 (`store_{8,16,32,64}()` and `load_{8,16,32,64}()`) for storing/loading data
159 to/from non-atomic shared memory.
161 CDSChecker can also check boolean assertions in your test programs. Just
162 include `<model-assert.h>` and use the `MODEL_ASSERT()` macro in your test program.
163 CDSChecker will report a bug in any possible execution in which the argument to
164 `MODEL_ASSERT()` evaluates to false (that is, 0).
166 Test programs should be compiled against our shared library (libmodel.so) using
167 the headers in the `include/` directory. Then the shared library must be made
168 available to the dynamic linker, using the `LD_LIBRARY_PATH` environment
169 variable, for instance.
172 ### Supported C11/C++11 APIs ###
174 To model-check multithreaded code properly, CDSChecker needs to instrument any
175 concurrency-related API calls made in your code. Currently, we support parts of
176 the following thread-support libraries. The C versions can be used in either C
179 * `<atomic>`, `<cstdatomic>`, `<stdatomic.h>`
180 * `<condition_variable>`
184 Because we want to extend support to legacy (i.e., non-C++11) compilers, we do
185 not support some new C++11 features that can't be implemented in C++03 (e.g.,
188 Reading an execution trace
189 --------------------------
191 When CDSChecker detects a bug in your program (or when run with the `--verbose`
192 flag), it prints the output of the program run (STDOUT) along with some summary
193 trace information for the execution in question. The trace is given as a
194 sequence of lines, where each line represents an operation in the execution
195 trace. These lines are ordered by the order in which they were run by CDSChecker
196 (i.e., the "execution order"), which does not necessarily align with the "order"
197 of the values observed (i.e., the modification order or the reads-from
200 The following list describes each of the columns in the execution trace output:
202 * \#: The sequence number within the execution. That is, sequence number "9"
203 means the operation was the 9th operation executed by CDSChecker. Note that
204 this represents the execution order, not necessarily any other order (e.g.,
205 modification order or reads-from).
207 * t: The thread number
209 * Action type: The type of operation performed
211 * MO: The memory-order for this operation (i.e., `memory_order_XXX`, where `XXX` is
212 `relaxed`, `release`, `acquire`, `rel_acq`, or `seq_cst`)
214 * Location: The memory location on which this operation is operating. This is
215 well-defined for atomic write/read/RMW, but other operations are subject to
216 CDSChecker implementation details.
218 * Value: For reads/writes/RMW, the value returned by the operation. Note that
219 for RMW, this is the value that is *read*, not the value that was *written*.
220 For other operations, 'value' may have some CDSChecker-internal meaning, or
221 it may simply be a don't-care (such as `0xdeadbeef`).
223 * Rf: For reads, the sequence number of the operation from which it reads.
224 [Note: If the execution is a partial, infeasible trace (labeled INFEASIBLE),
225 as printed during `--verbose` execution, reads may not be resolved and so may
226 have Rf=? or Rf=Px, where x is a promised future value.]
228 * CV: The clock vector, encapsulating the happens-before relation (see our
229 paper, or the C/C++ memory model itself). We use a Lamport-style clock vector
230 similar to [1]. The "clock" is just the sequence number (#). The clock vector
231 can be read as follows:
233 Each entry is indexed as CV[i], where
235 i = 0, 1, 2, ..., <number of threads>
237 So for any thread i, we say CV[i] is the sequence number of the most recent
238 operation in thread i such that operation i happens-before this operation.
239 Notably, thread 0 is reserved as a dummy thread for certain CDSChecker
242 See the following example trace:
244 ------------------------------------------------------------------------------------
245 # t Action type MO Location Value Rf CV
246 ------------------------------------------------------------------------------------
247 1 1 thread start seq_cst 0x7f68ff11e7c0 0xdeadbeef ( 0, 1)
248 2 1 init atomic relaxed 0x601068 0 ( 0, 2)
249 3 1 init atomic relaxed 0x60106c 0 ( 0, 3)
250 4 1 thread create seq_cst 0x7f68fe51c710 0x7f68fe51c6e0 ( 0, 4)
251 5 2 thread start seq_cst 0x7f68ff11ebc0 0xdeadbeef ( 0, 4, 5)
252 6 2 atomic read relaxed 0x60106c 0 3 ( 0, 4, 6)
253 7 1 thread create seq_cst 0x7f68fe51c720 0x7f68fe51c6e0 ( 0, 7)
254 8 3 thread start seq_cst 0x7f68ff11efc0 0xdeadbeef ( 0, 7, 0, 8)
255 9 2 atomic write relaxed 0x601068 0 ( 0, 4, 9)
256 10 3 atomic read relaxed 0x601068 0 2 ( 0, 7, 0, 10)
257 11 2 thread finish seq_cst 0x7f68ff11ebc0 0xdeadbeef ( 0, 4, 11)
258 12 3 atomic write relaxed 0x60106c 0x2a ( 0, 7, 0, 12)
259 13 1 thread join seq_cst 0x7f68ff11ebc0 0x2 ( 0, 13, 11)
260 14 3 thread finish seq_cst 0x7f68ff11efc0 0xdeadbeef ( 0, 7, 0, 14)
261 15 1 thread join seq_cst 0x7f68ff11efc0 0x3 ( 0, 15, 11, 14)
262 16 1 thread finish seq_cst 0x7f68ff11e7c0 0xdeadbeef ( 0, 16, 11, 14)
264 ------------------------------------------------------------------------------------
266 Now consider, for example, operation 10:
268 This is the 10th operation in the execution order. It is an atomic read-relaxed
269 operation performed by thread 3 at memory address `0x601068`. It reads the value
270 "0", which was written by the 2nd operation in the execution order. Its clock
271 vector consists of the following values:
273 CV[0] = 0, CV[1] = 7, CV[2] = 0, CV[3] = 10
275 End of Execution Summary
276 ------------------------
278 CDSChecker prints summary statistics at the end of each execution. These
279 summaries are based off of a few different properties of an execution, which we
280 will break down here:
282 * An _infeasible_ execution is an execution which is not consistent with the
283 memory model. Such an execution can be considered overhead for the
284 model-checker, since it should never appear in practice.
286 * A _buggy_ execution is an execution in which CDSChecker has found a real
287 bug: a data race, a deadlock, failure of a user-provided assertion, or an
288 uninitialized load, for instance. CDSChecker will only report bugs in feasible
291 * A _redundant_ execution is a feasible execution that is exploring the same
292 state space explored by a previous feasible execution. Such exploration is
293 another instance of overhead, so CDSChecker terminates these executions as
294 soon as they are detected. CDSChecker is mostly able to avoid such executions
295 but may encounter them if a fairness option is enabled.
297 Now, we can examine the end-of-execution summary of one test program:
299 $ ./run.sh test/rmwprog.o
301 ******* Model-checking complete: *******
302 Number of complete, bug-free executions: 6
303 Number of redundant executions: 0
304 Number of buggy executions: 0
305 Number of infeasible executions: 29
308 * _Number of complete, bug-free executions:_ these are feasible, non-buggy, and
309 non-redundant executions. They each represent different, legal behaviors you
310 can expect to see in practice.
312 * _Number of redundant executions:_ these are feasible but redundant executions
313 that were terminated as soon as CDSChecker noticed the redundancy.
315 * _Number of buggy executions:_ these are feasible, buggy executions. These are
316 the trouble spots where your program is triggering a bug or assertion.
317 Ideally, this number should be 0.
319 * _Number of infeasible executions:_ these are infeasible executions,
320 representing some of the overhead of model-checking.
322 * _Total executions:_ the total number of executions explored by CDSChecker.
323 Should be the sum of the above categories, since they are mutually exclusive.
326 Other Notes and Pitfalls
327 ------------------------
329 * Many programs require some form of fairness in order to terminate in a finite
330 amount of time. CDSChecker supports the `-y num` and `-f num` flags for these
331 cases. The `-y` option (yield-based fairness) is preferable, but it requires
332 careful usage of yields (i.e., `thrd_yield()`) in the test program. For
333 programs without proper `thrd_yield()`, you may consider using `-f` instead.
335 * Deadlock detection: CDSChecker can detect deadlocks. For instance, try the
336 following test program.
338 > ./run.sh test/deadlock.o
340 Deadlock detection currently detects when a thread is about to step into a
341 deadlock, without actually including the final step in the trace. But you can
342 examine the program to see the next step.
344 * CDSChecker has to speculatively explore many execution behaviors due to the
345 relaxed memory model, and many of these turn out to be infeasible (that is,
346 they cannot be legally produced by the memory model). CDSChecker discards
347 these executions as soon as it identifies them (see the "Number of infeasible
348 executions" statistic); however, the speculation can occasionally cause
349 CDSChecker to hit unexpected parts of the unit test program (causing a
350 division by 0, for instance). In such programs, you might consider running
351 CDSChecker with the `-u num` option.
353 * Related to the previous point, CDSChecker may report more than one bug for a
354 particular candidate execution. This is because some bugs may not be
355 reportable until CDSChecker has explored more of the program, and in the
356 time between initial discovery and final assessment of the bug, CDSChecker may
357 discover another bug.
359 * Data races may be reported as multiple bugs, one for each byte-address of the
360 data race in question. See, for example, this run:
362 $ ./run.sh test/releaseseq.o
364 Bug report: 4 bugs detected
365 [BUG] Data race detected @ address 0x601078:
366 Access 1: write in thread 2 @ clock 4
367 Access 2: read in thread 3 @ clock 9
368 [BUG] Data race detected @ address 0x601079:
369 Access 1: write in thread 2 @ clock 4
370 Access 2: read in thread 3 @ clock 9
371 [BUG] Data race detected @ address 0x60107a:
372 Access 1: write in thread 2 @ clock 4
373 Access 2: read in thread 3 @ clock 9
374 [BUG] Data race detected @ address 0x60107b:
375 Access 1: write in thread 2 @ clock 4
376 Access 2: read in thread 3 @ clock 9
382 The CDSChecker project page:
384 > <http://demsky.eecs.uci.edu/c11modelchecker.php>
386 The CDSChecker source and accompanying benchmarks on Gitweb:
388 > <http://demsky.eecs.uci.edu/git/?p=model-checker.git>
390 > <http://demsky.eecs.uci.edu/git/?p=model-checker-benchmarks.git>
396 Please feel free to contact us for more information. Bug reports are welcome,
397 and we are happy to hear from our users. We are also very interested to know if
398 CDSChecker catches bugs in your programs.
400 Contact Brian Norris at <banorris@uci.edu> or Brian Demsky at <bdemsky@uci.edu>.
406 [1] L. Lamport. Time, clocks, and the ordering of events in a distributed
407 system. CACM, 21(7):558-565, July 1978.