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/):
42 git clone git://demsky.eecs.uci.edu/model-checker.git
44 Source code can also be downloaded via the snapshot links on Gitweb (found in
45 the __See Also__ section).
47 Get the benchmarks (not required; distributed separately):
50 git clone git://demsky.eecs.uci.edu/model-checker-benchmarks.git benchmarks
52 Compile the model checker:
56 Compile the benchmarks:
60 Run a simple example (the `run.sh` script does some very minimal processing for
63 ./run.sh test/userprog.o
65 To see the help message on how to run CDSChecker, execute:
75 > Controls the liveness of the memory system. Note that multithreaded programs
76 > often rely on memory liveness for termination, so this parameter is
77 > necessary for such programs.
79 > Liveness is controlled by `num`: the number of times a load is allowed to
80 > see the same store when a newer store exists---one that is ordered later in
81 > the modification order.
85 > Turns on CHESS-like yield-based fairness support (requires `thrd_yield()`
86 > instrumentation in test program).
90 > Turns on alternative fairness support (less desirable than `-y`). A
91 > necessary alternative for some programs that do not support yield-based
96 > Verbose: show all executions and not just buggy ones.
100 > Constrain how long we will run to wait for a future value past when it is
105 > Value to provide to atomics loads from uninitialized memory locations. The
106 > default is 0, but this may cause some programs to throw exceptions
107 > (segfault) before the model checker prints a trace.
121 Many simple tests are located in the `tests/` directory. You may also want to
122 try the larger benchmarks (distributed separately), which can be placed under
123 the `benchmarks/` directory. After building CDSChecker, you can build and run
124 the benchmarks as follows:
129 > # run barrier test with fairness/memory liveness
130 > ./run.sh barrier/barrier -y -m 2
132 > # Linux reader/write lock test with fairness/memory liveness
133 > ./run.sh linuxrwlocks/linuxrwlocks -y -m 2
135 > # run all benchmarks and provide timing results
139 Running your own code
140 ---------------------
142 You likely want to test your own code, not just our simple tests. To do so, you
143 need to perform a few steps.
145 First, because CDSChecker executes your program dozens (if not hundreds or
146 thousands) of times, you will have the most success if your code is written as a
147 unit test and not as a full-blown program.
149 Second, because CDSChecker must be able to manage your program for you, your
150 program should declare its main entry point as `user_main(int, char**)` rather
151 than `main(int, char**)`.
153 Third, test programs must use the standard C11/C++11 library headers (see below
154 for supported APIs) and must compile against the versions provided in
155 CDSChecker's `include/` directory. Notably, we only support C11 thread syntax
156 (`thrd_t`, etc. from `<thread.h>`).
158 Test programs may also use our included happens-before race detector by
159 including <librace.h> and utilizing the appropriate functions
160 (`store_{8,16,32,64}()` and `load_{8,16,32,64}()`) for storing/loading data
161 to/from non-atomic shared memory.
163 CDSChecker can also check boolean assertions in your test programs. Just
164 include `<model-assert.h>` and use the `MODEL_ASSERT()` macro in your test program.
165 CDSChecker will report a bug in any possible execution in which the argument to
166 `MODEL_ASSERT()` evaluates to false (that is, 0).
168 Test programs should be compiled against our shared library (libmodel.so) using
169 the headers in the `include/` directory. Then the shared library must be made
170 available to the dynamic linker, using the `LD_LIBRARY_PATH` environment
171 variable, for instance.
174 ### Supported C11/C++11 APIs ###
176 To model-check multithreaded code properly, CDSChecker needs to instrument any
177 concurrency-related API calls made in your code. Currently, we support parts of
178 the following thread-support libraries. The C versions can be used in either C
181 * `<atomic>`, `<cstdatomic>`, `<stdatomic.h>`
182 * `<condition_variable>`
186 Because we want to extend support to legacy (i.e., non-C++11) compilers, we do
187 not support some new C++11 features that can't be implemented in C++03 (e.g.,
190 Reading an execution trace
191 --------------------------
193 When CDSChecker detects a bug in your program (or when run with the `--verbose`
194 flag), it prints the output of the program run (STDOUT) along with some summary
195 trace information for the execution in question. The trace is given as a
196 sequence of lines, where each line represents an operation in the execution
197 trace. These lines are ordered by the order in which they were run by CDSChecker
198 (i.e., the "execution order"), which does not necessarily align with the "order"
199 of the values observed (i.e., the modification order or the reads-from
202 The following list describes each of the columns in the execution trace output:
204 * \#: The sequence number within the execution. That is, sequence number "9"
205 means the operation was the 9th operation executed by CDSChecker. Note that
206 this represents the execution order, not necessarily any other order (e.g.,
207 modification order or reads-from).
209 * t: The thread number
211 * Action type: The type of operation performed
213 * MO: The memory-order for this operation (i.e., `memory_order_XXX`, where `XXX` is
214 `relaxed`, `release`, `acquire`, `rel_acq`, or `seq_cst`)
216 * Location: The memory location on which this operation is operating. This is
217 well-defined for atomic write/read/RMW, but other operations are subject to
218 CDSChecker implementation details.
220 * Value: For reads/writes/RMW, the value returned by the operation. Note that
221 for RMW, this is the value that is *read*, not the value that was *written*.
222 For other operations, 'value' may have some CDSChecker-internal meaning, or
223 it may simply be a don't-care (such as `0xdeadbeef`).
225 * Rf: For reads, the sequence number of the operation from which it reads.
226 [Note: If the execution is a partial, infeasible trace (labeled INFEASIBLE),
227 as printed during `--verbose` execution, reads may not be resolved and so may
228 have Rf=? or Rf=Px, where x is a promised future value.]
230 * CV: The clock vector, encapsulating the happens-before relation (see our
231 paper, or the C/C++ memory model itself). We use a Lamport-style clock vector
232 similar to [1]. The "clock" is just the sequence number (#). The clock vector
233 can be read as follows:
235 Each entry is indexed as CV[i], where
237 i = 0, 1, 2, ..., <number of threads>
239 So for any thread i, we say CV[i] is the sequence number of the most recent
240 operation in thread i such that operation i happens-before this operation.
241 Notably, thread 0 is reserved as a dummy thread for certain CDSChecker
244 See the following example trace:
246 ------------------------------------------------------------------------------------
247 # t Action type MO Location Value Rf CV
248 ------------------------------------------------------------------------------------
249 1 1 thread start seq_cst 0x7f68ff11e7c0 0xdeadbeef ( 0, 1)
250 2 1 init atomic relaxed 0x601068 0 ( 0, 2)
251 3 1 init atomic relaxed 0x60106c 0 ( 0, 3)
252 4 1 thread create seq_cst 0x7f68fe51c710 0x7f68fe51c6e0 ( 0, 4)
253 5 2 thread start seq_cst 0x7f68ff11ebc0 0xdeadbeef ( 0, 4, 5)
254 6 2 atomic read relaxed 0x60106c 0 3 ( 0, 4, 6)
255 7 1 thread create seq_cst 0x7f68fe51c720 0x7f68fe51c6e0 ( 0, 7)
256 8 3 thread start seq_cst 0x7f68ff11efc0 0xdeadbeef ( 0, 7, 0, 8)
257 9 2 atomic write relaxed 0x601068 0 ( 0, 4, 9)
258 10 3 atomic read relaxed 0x601068 0 2 ( 0, 7, 0, 10)
259 11 2 thread finish seq_cst 0x7f68ff11ebc0 0xdeadbeef ( 0, 4, 11)
260 12 3 atomic write relaxed 0x60106c 0x2a ( 0, 7, 0, 12)
261 13 1 thread join seq_cst 0x7f68ff11ebc0 0x2 ( 0, 13, 11)
262 14 3 thread finish seq_cst 0x7f68ff11efc0 0xdeadbeef ( 0, 7, 0, 14)
263 15 1 thread join seq_cst 0x7f68ff11efc0 0x3 ( 0, 15, 11, 14)
264 16 1 thread finish seq_cst 0x7f68ff11e7c0 0xdeadbeef ( 0, 16, 11, 14)
266 ------------------------------------------------------------------------------------
268 Now consider, for example, operation 10:
270 This is the 10th operation in the execution order. It is an atomic read-relaxed
271 operation performed by thread 3 at memory address `0x601068`. It reads the value
272 "0", which was written by the 2nd operation in the execution order. Its clock
273 vector consists of the following values:
275 CV[0] = 0, CV[1] = 7, CV[2] = 0, CV[3] = 10
277 End of Execution Summary
278 ------------------------
280 CDSChecker prints summary statistics at the end of each execution. These
281 summaries are based off of a few different properties of an execution, which we
282 will break down here:
284 * An _infeasible_ execution is an execution which is not consistent with the
285 memory model. Such an execution can be considered overhead for the
286 model-checker, since it should never appear in practice.
288 * A _buggy_ execution is an execution in which CDSChecker has found a real
289 bug: a data race, a deadlock, failure of a user-provided assertion, or an
290 uninitialized load, for instance. CDSChecker will only report bugs in feasible
293 * A _redundant_ execution is a feasible execution that is exploring the same
294 state space explored by a previous feasible execution. Such exploration is
295 another instance of overhead, so CDSChecker terminates these executions as
296 soon as they are detected. CDSChecker is mostly able to avoid such executions
297 but may encounter them if a fairness option is enabled.
299 Now, we can examine the end-of-execution summary of one test program:
301 $ ./run.sh test/rmwprog.o
303 ******* Model-checking complete: *******
304 Number of complete, bug-free executions: 6
305 Number of redundant executions: 0
306 Number of buggy executions: 0
307 Number of infeasible executions: 29
310 * _Number of complete, bug-free executions:_ these are feasible, non-buggy, and
311 non-redundant executions. They each represent different, legal behaviors you
312 can expect to see in practice.
314 * _Number of redundant executions:_ these are feasible but redundant executions
315 that were terminated as soon as CDSChecker noticed the redundancy.
317 * _Number of buggy executions:_ these are feasible, buggy executions. These are
318 the trouble spots where your program is triggering a bug or assertion.
319 Ideally, this number should be 0.
321 * _Number of infeasible executions:_ these are infeasible executions,
322 representing some of the overhead of model-checking.
324 * _Total executions:_ the total number of executions explored by CDSChecker.
325 Should be the sum of the above categories, since they are mutually exclusive.
328 Other Notes and Pitfalls
329 ------------------------
331 * Many programs require some form of fairness in order to terminate in a finite
332 amount of time. CDSChecker supports the `-y num` and `-f num` flags for these
333 cases. The `-y` option (yield-based fairness) is preferable, but it requires
334 careful usage of yields (i.e., `thrd_yield()`) in the test program. For
335 programs without proper `thrd_yield()`, you may consider using `-f` instead.
337 * Deadlock detection: CDSChecker can detect deadlocks. For instance, try the
338 following test program.
340 > ./run.sh test/deadlock.o
342 Deadlock detection currently detects when a thread is about to step into a
343 deadlock, without actually including the final step in the trace. But you can
344 examine the program to see the next step.
346 * CDSChecker has to speculatively explore many execution behaviors due to the
347 relaxed memory model, and many of these turn out to be infeasible (that is,
348 they cannot be legally produced by the memory model). CDSChecker discards
349 these executions as soon as it identifies them (see the "Number of infeasible
350 executions" statistic); however, the speculation can occasionally cause
351 CDSChecker to hit unexpected parts of the unit test program (causing a
352 division by 0, for instance). In such programs, you might consider running
353 CDSChecker with the `-u num` option.
355 * Related to the previous point, CDSChecker may report more than one bug for a
356 particular candidate execution. This is because some bugs may not be
357 reportable until CDSChecker has explored more of the program, and in the
358 time between initial discovery and final assessment of the bug, CDSChecker may
359 discover another bug.
361 * Data races may be reported as multiple bugs, one for each byte-address of the
362 data race in question. See, for example, this run:
364 $ ./run.sh test/releaseseq.o
366 Bug report: 4 bugs detected
367 [BUG] Data race detected @ address 0x601078:
368 Access 1: write in thread 2 @ clock 4
369 Access 2: read in thread 3 @ clock 9
370 [BUG] Data race detected @ address 0x601079:
371 Access 1: write in thread 2 @ clock 4
372 Access 2: read in thread 3 @ clock 9
373 [BUG] Data race detected @ address 0x60107a:
374 Access 1: write in thread 2 @ clock 4
375 Access 2: read in thread 3 @ clock 9
376 [BUG] Data race detected @ address 0x60107b:
377 Access 1: write in thread 2 @ clock 4
378 Access 2: read in thread 3 @ clock 9
384 The CDSChecker project page:
386 > <http://demsky.eecs.uci.edu/c11modelchecker.php>
388 The CDSChecker source and accompanying benchmarks on Gitweb:
390 > <http://demsky.eecs.uci.edu/git/?p=model-checker.git>
392 > <http://demsky.eecs.uci.edu/git/?p=model-checker-benchmarks.git>
398 Please feel free to contact us for more information. Bug reports are welcome,
399 and we are happy to hear from our users. We are also very interested to know if
400 CDSChecker catches bugs in your programs.
402 Contact Brian Norris at <banorris@uci.edu> or Brian Demsky at <bdemsky@uci.edu>.
408 [1] L. Lamport. Time, clocks, and the ordering of events in a distributed
409 system. CACM, 21(7):558-565, July 1978.