1 RCU Torture Test Operation
4 CONFIG_RCU_TORTURE_TEST
6 The CONFIG_RCU_TORTURE_TEST config option is available for all RCU
7 implementations. It creates an rcutorture kernel module that can
8 be loaded to run a torture test. The test periodically outputs
9 status messages via printk(), which can be examined via the dmesg
10 command (perhaps grepping for "torture"). The test is started
11 when the module is loaded, and stops when the module is unloaded.
16 This module has the following parameters:
18 fqs_duration Duration (in microseconds) of artificially induced bursts
19 of force_quiescent_state() invocations. In RCU
20 implementations having force_quiescent_state(), these
21 bursts help force races between forcing a given grace
22 period and that grace period ending on its own.
24 fqs_holdoff Holdoff time (in microseconds) between consecutive calls
25 to force_quiescent_state() within a burst.
27 fqs_stutter Wait time (in seconds) between consecutive bursts
28 of calls to force_quiescent_state().
30 gp_normal Make the fake writers use normal synchronous grace-period
33 gp_exp Make the fake writers use expedited synchronous grace-period
34 primitives. If both gp_normal and gp_exp are set, or
35 if neither gp_normal nor gp_exp are set, then randomly
36 choose the primitive so that about 50% are normal and
37 50% expedited. By default, neither are set, which
38 gives best overall test coverage.
40 irqreader Says to invoke RCU readers from irq level. This is currently
41 done via timers. Defaults to "1" for variants of RCU that
42 permit this. (Or, more accurately, variants of RCU that do
43 -not- permit this know to ignore this variable.)
45 n_barrier_cbs If this is nonzero, RCU barrier testing will be conducted,
46 in which case n_barrier_cbs specifies the number of
47 RCU callbacks (and corresponding kthreads) to use for
48 this testing. The value cannot be negative. If you
49 specify this to be non-zero when torture_type indicates a
50 synchronous RCU implementation (one for which a member of
51 the synchronize_rcu() rather than the call_rcu() family is
52 used -- see the documentation for torture_type below), an
53 error will be reported and no testing will be carried out.
55 nfakewriters This is the number of RCU fake writer threads to run. Fake
56 writer threads repeatedly use the synchronous "wait for
57 current readers" function of the interface selected by
58 torture_type, with a delay between calls to allow for various
59 different numbers of writers running in parallel.
60 nfakewriters defaults to 4, which provides enough parallelism
61 to trigger special cases caused by multiple writers, such as
62 the synchronize_srcu() early return optimization.
64 nreaders This is the number of RCU reading threads supported.
65 The default is twice the number of CPUs. Why twice?
66 To properly exercise RCU implementations with preemptible
67 read-side critical sections.
70 The number of seconds between each attempt to execute a
71 randomly selected CPU-hotplug operation. Defaults to
72 zero, which disables CPU hotplugging. In HOTPLUG_CPU=n
73 kernels, rcutorture will silently refuse to do any
74 CPU-hotplug operations regardless of what value is
75 specified for onoff_interval.
77 onoff_holdoff The number of seconds to wait until starting CPU-hotplug
78 operations. This would normally only be used when
79 rcutorture was built into the kernel and started
80 automatically at boot time, in which case it is useful
81 in order to avoid confusing boot-time code with CPUs
85 The number of seconds to keep the test threads affinitied
86 to a particular subset of the CPUs, defaults to 3 seconds.
87 Used in conjunction with test_no_idle_hz.
89 shutdown_secs The number of seconds to run the test before terminating
90 the test and powering off the system. The default is
91 zero, which disables test termination and system shutdown.
92 This capability is useful for automated testing.
94 stall_cpu The number of seconds that a CPU should be stalled while
95 within both an rcu_read_lock() and a preempt_disable().
96 This stall happens only once per rcutorture run.
97 If you need multiple stalls, use modprobe and rmmod to
98 repeatedly run rcutorture. The default for stall_cpu
99 is zero, which prevents rcutorture from stalling a CPU.
101 Note that attempts to rmmod rcutorture while the stall
102 is ongoing will hang, so be careful what value you
103 choose for this module parameter! In addition, too-large
104 values for stall_cpu might well induce failures and
105 warnings in other parts of the kernel. You have been
109 The number of seconds to wait after rcutorture starts
110 before stalling a CPU. Defaults to 10 seconds.
112 stat_interval The number of seconds between output of torture
113 statistics (via printk()). Regardless of the interval,
114 statistics are printed when the module is unloaded.
115 Setting the interval to zero causes the statistics to
116 be printed -only- when the module is unloaded, and this
119 stutter The length of time to run the test before pausing for this
120 same period of time. Defaults to "stutter=5", so as
121 to run and pause for (roughly) five-second intervals.
122 Specifying "stutter=0" causes the test to run continuously
123 without pausing, which is the old default behavior.
125 test_boost Whether or not to test the ability of RCU to do priority
126 boosting. Defaults to "test_boost=1", which performs
127 RCU priority-inversion testing only if the selected
128 RCU implementation supports priority boosting. Specifying
129 "test_boost=0" never performs RCU priority-inversion
130 testing. Specifying "test_boost=2" performs RCU
131 priority-inversion testing even if the selected RCU
132 implementation does not support RCU priority boosting,
133 which can be used to test rcutorture's ability to
134 carry out RCU priority-inversion testing.
137 The number of seconds in an RCU priority-inversion test
138 cycle. Defaults to "test_boost_interval=7". It is
139 usually wise for this value to be relatively prime to
140 the value selected for "stutter".
143 The number of seconds to do RCU priority-inversion testing
144 within any given "test_boost_interval". Defaults to
145 "test_boost_duration=4".
147 test_no_idle_hz Whether or not to test the ability of RCU to operate in
148 a kernel that disables the scheduling-clock interrupt to
149 idle CPUs. Boolean parameter, "1" to test, "0" otherwise.
150 Defaults to omitting this test.
152 torture_type The type of RCU to test, with string values as follows:
154 "rcu": rcu_read_lock(), rcu_read_unlock() and call_rcu(),
155 along with expedited, synchronous, and polling
158 "rcu_bh": rcu_read_lock_bh(), rcu_read_unlock_bh(), and
159 call_rcu_bh(), along with expedited and synchronous
162 "rcu_busted": This tests an intentionally incorrect version
163 of RCU in order to help test rcutorture itself.
165 "srcu": srcu_read_lock(), srcu_read_unlock() and
166 call_srcu(), along with expedited and
167 synchronous variants.
169 "sched": preempt_disable(), preempt_enable(), and
170 call_rcu_sched(), along with expedited,
171 synchronous, and polling variants.
173 "tasks": voluntary context switch and call_rcu_tasks(),
174 along with expedited and synchronous variants.
178 verbose Enable debug printk()s. Default is disabled.
183 The statistics output is as follows:
185 rcu-torture:--- Start of test: nreaders=16 nfakewriters=4 stat_interval=30 verbose=0 test_no_idle_hz=1 shuffle_interval=3 stutter=5 irqreader=1 fqs_duration=0 fqs_holdoff=0 fqs_stutter=3 test_boost=1/0 test_boost_interval=7 test_boost_duration=4
186 rcu-torture: rtc: (null) ver: 155441 tfle: 0 rta: 155441 rtaf: 8884 rtf: 155440 rtmbe: 0 rtbe: 0 rtbke: 0 rtbre: 0 rtbf: 0 rtb: 0 nt: 3055767
187 rcu-torture: Reader Pipe: 727860534 34213 0 0 0 0 0 0 0 0 0
188 rcu-torture: Reader Batch: 727877838 17003 0 0 0 0 0 0 0 0 0
189 rcu-torture: Free-Block Circulation: 155440 155440 155440 155440 155440 155440 155440 155440 155440 155440 0
190 rcu-torture:--- End of test: SUCCESS: nreaders=16 nfakewriters=4 stat_interval=30 verbose=0 test_no_idle_hz=1 shuffle_interval=3 stutter=5 irqreader=1 fqs_duration=0 fqs_holdoff=0 fqs_stutter=3 test_boost=1/0 test_boost_interval=7 test_boost_duration=4
192 The command "dmesg | grep torture:" will extract this information on
193 most systems. On more esoteric configurations, it may be necessary to
194 use other commands to access the output of the printk()s used by
195 the RCU torture test. The printk()s use KERN_ALERT, so they should
198 The first and last lines show the rcutorture module parameters, and the
199 last line shows either "SUCCESS" or "FAILURE", based on rcutorture's
200 automatic determination as to whether RCU operated correctly.
202 The entries are as follows:
204 o "rtc": The hexadecimal address of the structure currently visible
207 o "ver": The number of times since boot that the RCU writer task
208 has changed the structure visible to readers.
210 o "tfle": If non-zero, indicates that the "torture freelist"
211 containing structures to be placed into the "rtc" area is empty.
212 This condition is important, since it can fool you into thinking
213 that RCU is working when it is not. :-/
215 o "rta": Number of structures allocated from the torture freelist.
217 o "rtaf": Number of allocations from the torture freelist that have
218 failed due to the list being empty. It is not unusual for this
219 to be non-zero, but it is bad for it to be a large fraction of
220 the value indicated by "rta".
222 o "rtf": Number of frees into the torture freelist.
224 o "rtmbe": A non-zero value indicates that rcutorture believes that
225 rcu_assign_pointer() and rcu_dereference() are not working
226 correctly. This value should be zero.
228 o "rtbe": A non-zero value indicates that one of the rcu_barrier()
229 family of functions is not working correctly.
231 o "rtbke": rcutorture was unable to create the real-time kthreads
232 used to force RCU priority inversion. This value should be zero.
234 o "rtbre": Although rcutorture successfully created the kthreads
235 used to force RCU priority inversion, it was unable to set them
236 to the real-time priority level of 1. This value should be zero.
238 o "rtbf": The number of times that RCU priority boosting failed
239 to resolve RCU priority inversion.
241 o "rtb": The number of times that rcutorture attempted to force
242 an RCU priority inversion condition. If you are testing RCU
243 priority boosting via the "test_boost" module parameter, this
244 value should be non-zero.
246 o "nt": The number of times rcutorture ran RCU read-side code from
247 within a timer handler. This value should be non-zero only
248 if you specified the "irqreader" module parameter.
250 o "Reader Pipe": Histogram of "ages" of structures seen by readers.
251 If any entries past the first two are non-zero, RCU is broken.
252 And rcutorture prints the error flag string "!!!" to make sure
253 you notice. The age of a newly allocated structure is zero,
254 it becomes one when removed from reader visibility, and is
255 incremented once per grace period subsequently -- and is freed
256 after passing through (RCU_TORTURE_PIPE_LEN-2) grace periods.
258 The output displayed above was taken from a correctly working
259 RCU. If you want to see what it looks like when broken, break
262 o "Reader Batch": Another histogram of "ages" of structures seen
263 by readers, but in terms of counter flips (or batches) rather
264 than in terms of grace periods. The legal number of non-zero
265 entries is again two. The reason for this separate view is that
266 it is sometimes easier to get the third entry to show up in the
267 "Reader Batch" list than in the "Reader Pipe" list.
269 o "Free-Block Circulation": Shows the number of torture structures
270 that have reached a given point in the pipeline. The first element
271 should closely correspond to the number of structures allocated,
272 the second to the number that have been removed from reader view,
273 and all but the last remaining to the corresponding number of
274 passes through a grace period. The last entry should be zero,
275 as it is only incremented if a torture structure's counter
276 somehow gets incremented farther than it should.
278 Different implementations of RCU can provide implementation-specific
279 additional information. For example, Tree SRCU provides the following
282 srcud-torture: Tree SRCU per-CPU(idx=0): 0(35,-21) 1(-4,24) 2(1,1) 3(-26,20) 4(28,-47) 5(-9,4) 6(-10,14) 7(-14,11) T(1,6)
284 This line shows the per-CPU counter state, in this case for Tree SRCU
285 using a dynamically allocated srcu_struct (hence "srcud-" rather than
286 "srcu-"). The numbers in parentheses are the values of the "old" and
287 "current" counters for the corresponding CPU. The "idx" value maps the
288 "old" and "current" values to the underlying array, and is useful for
289 debugging. The final "T" entry contains the totals of the counters.
294 The following script may be used to torture RCU:
301 dmesg | grep torture:
303 The output can be manually inspected for the error flag of "!!!".
304 One could of course create a more elaborate script that automatically
305 checked for such errors. The "rmmod" command forces a "SUCCESS",
306 "FAILURE", or "RCU_HOTPLUG" indication to be printk()ed. The first
307 two are self-explanatory, while the last indicates that while there
308 were no RCU failures, CPU-hotplug problems were detected.
310 However, the tools/testing/selftests/rcutorture/bin/kvm.sh script
311 provides better automation, including automatic failure analysis.
312 It assumes a qemu/kvm-enabled platform, and runs guest OSes out of initrd.
313 See tools/testing/selftests/rcutorture/doc/initrd.txt for instructions
314 on setting up such an initrd.