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.
13 CONFIG_RCU_TORTURE_TEST_RUNNABLE
15 It is also possible to specify CONFIG_RCU_TORTURE_TEST=y, which will
16 result in the tests being loaded into the base kernel. In this case,
17 the CONFIG_RCU_TORTURE_TEST_RUNNABLE config option is used to specify
18 whether the RCU torture tests are to be started immediately during
19 boot or whether the /proc/sys/kernel/rcutorture_runnable file is used
20 to enable them. This /proc file can be used to repeatedly pause and
21 restart the tests, regardless of the initial state specified by the
22 CONFIG_RCU_TORTURE_TEST_RUNNABLE config option.
24 You will normally -not- want to start the RCU torture tests during boot
25 (and thus the default is CONFIG_RCU_TORTURE_TEST_RUNNABLE=n), but doing
26 this can sometimes be useful in finding boot-time bugs.
31 This module has the following parameters:
33 fqs_duration Duration (in microseconds) of artificially induced bursts
34 of force_quiescent_state() invocations. In RCU
35 implementations having force_quiescent_state(), these
36 bursts help force races between forcing a given grace
37 period and that grace period ending on its own.
39 fqs_holdoff Holdoff time (in microseconds) between consecutive calls
40 to force_quiescent_state() within a burst.
42 fqs_stutter Wait time (in seconds) between consecutive bursts
43 of calls to force_quiescent_state().
45 gp_normal Make the fake writers use normal synchronous grace-period
48 gp_exp Make the fake writers use expedited synchronous grace-period
49 primitives. If both gp_normal and gp_exp are set, or
50 if neither gp_normal nor gp_exp are set, then randomly
51 choose the primitive so that about 50% are normal and
52 50% expedited. By default, neither are set, which
53 gives best overall test coverage.
55 irqreader Says to invoke RCU readers from irq level. This is currently
56 done via timers. Defaults to "1" for variants of RCU that
57 permit this. (Or, more accurately, variants of RCU that do
58 -not- permit this know to ignore this variable.)
60 n_barrier_cbs If this is nonzero, RCU barrier testing will be conducted,
61 in which case n_barrier_cbs specifies the number of
62 RCU callbacks (and corresponding kthreads) to use for
63 this testing. The value cannot be negative. If you
64 specify this to be non-zero when torture_type indicates a
65 synchronous RCU implementation (one for which a member of
66 the synchronize_rcu() rather than the call_rcu() family is
67 used -- see the documentation for torture_type below), an
68 error will be reported and no testing will be carried out.
70 nfakewriters This is the number of RCU fake writer threads to run. Fake
71 writer threads repeatedly use the synchronous "wait for
72 current readers" function of the interface selected by
73 torture_type, with a delay between calls to allow for various
74 different numbers of writers running in parallel.
75 nfakewriters defaults to 4, which provides enough parallelism
76 to trigger special cases caused by multiple writers, such as
77 the synchronize_srcu() early return optimization.
79 nreaders This is the number of RCU reading threads supported.
80 The default is twice the number of CPUs. Why twice?
81 To properly exercise RCU implementations with preemptible
82 read-side critical sections.
85 The number of seconds between each attempt to execute a
86 randomly selected CPU-hotplug operation. Defaults to
87 zero, which disables CPU hotplugging. In HOTPLUG_CPU=n
88 kernels, rcutorture will silently refuse to do any
89 CPU-hotplug operations regardless of what value is
90 specified for onoff_interval.
92 onoff_holdoff The number of seconds to wait until starting CPU-hotplug
93 operations. This would normally only be used when
94 rcutorture was built into the kernel and started
95 automatically at boot time, in which case it is useful
96 in order to avoid confusing boot-time code with CPUs
100 The number of seconds to keep the test threads affinitied
101 to a particular subset of the CPUs, defaults to 3 seconds.
102 Used in conjunction with test_no_idle_hz.
104 shutdown_secs The number of seconds to run the test before terminating
105 the test and powering off the system. The default is
106 zero, which disables test termination and system shutdown.
107 This capability is useful for automated testing.
109 stall_cpu The number of seconds that a CPU should be stalled while
110 within both an rcu_read_lock() and a preempt_disable().
111 This stall happens only once per rcutorture run.
112 If you need multiple stalls, use modprobe and rmmod to
113 repeatedly run rcutorture. The default for stall_cpu
114 is zero, which prevents rcutorture from stalling a CPU.
116 Note that attempts to rmmod rcutorture while the stall
117 is ongoing will hang, so be careful what value you
118 choose for this module parameter! In addition, too-large
119 values for stall_cpu might well induce failures and
120 warnings in other parts of the kernel. You have been
124 The number of seconds to wait after rcutorture starts
125 before stalling a CPU. Defaults to 10 seconds.
127 stat_interval The number of seconds between output of torture
128 statistics (via printk()). Regardless of the interval,
129 statistics are printed when the module is unloaded.
130 Setting the interval to zero causes the statistics to
131 be printed -only- when the module is unloaded, and this
134 stutter The length of time to run the test before pausing for this
135 same period of time. Defaults to "stutter=5", so as
136 to run and pause for (roughly) five-second intervals.
137 Specifying "stutter=0" causes the test to run continuously
138 without pausing, which is the old default behavior.
140 test_boost Whether or not to test the ability of RCU to do priority
141 boosting. Defaults to "test_boost=1", which performs
142 RCU priority-inversion testing only if the selected
143 RCU implementation supports priority boosting. Specifying
144 "test_boost=0" never performs RCU priority-inversion
145 testing. Specifying "test_boost=2" performs RCU
146 priority-inversion testing even if the selected RCU
147 implementation does not support RCU priority boosting,
148 which can be used to test rcutorture's ability to
149 carry out RCU priority-inversion testing.
152 The number of seconds in an RCU priority-inversion test
153 cycle. Defaults to "test_boost_interval=7". It is
154 usually wise for this value to be relatively prime to
155 the value selected for "stutter".
158 The number of seconds to do RCU priority-inversion testing
159 within any given "test_boost_interval". Defaults to
160 "test_boost_duration=4".
162 test_no_idle_hz Whether or not to test the ability of RCU to operate in
163 a kernel that disables the scheduling-clock interrupt to
164 idle CPUs. Boolean parameter, "1" to test, "0" otherwise.
165 Defaults to omitting this test.
167 torture_type The type of RCU to test, with string values as follows:
169 "rcu": rcu_read_lock(), rcu_read_unlock() and call_rcu(),
170 along with expedited, synchronous, and polling
173 "rcu_bh": rcu_read_lock_bh(), rcu_read_unlock_bh(), and
174 call_rcu_bh(), along with expedited and synchronous
177 "rcu_busted": This tests an intentionally incorrect version
178 of RCU in order to help test rcutorture itself.
180 "srcu": srcu_read_lock(), srcu_read_unlock() and
181 call_srcu(), along with expedited and
182 synchronous variants.
184 "sched": preempt_disable(), preempt_enable(), and
185 call_rcu_sched(), along with expedited,
186 synchronous, and polling variants.
188 "tasks": voluntary context switch and call_rcu_tasks(),
189 along with expedited and synchronous variants.
193 verbose Enable debug printk()s. Default is disabled.
198 The statistics output is as follows:
200 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
201 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
202 rcu-torture: Reader Pipe: 727860534 34213 0 0 0 0 0 0 0 0 0
203 rcu-torture: Reader Batch: 727877838 17003 0 0 0 0 0 0 0 0 0
204 rcu-torture: Free-Block Circulation: 155440 155440 155440 155440 155440 155440 155440 155440 155440 155440 0
205 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
207 The command "dmesg | grep torture:" will extract this information on
208 most systems. On more esoteric configurations, it may be necessary to
209 use other commands to access the output of the printk()s used by
210 the RCU torture test. The printk()s use KERN_ALERT, so they should
213 The first and last lines show the rcutorture module parameters, and the
214 last line shows either "SUCCESS" or "FAILURE", based on rcutorture's
215 automatic determination as to whether RCU operated correctly.
217 The entries are as follows:
219 o "rtc": The hexadecimal address of the structure currently visible
222 o "ver": The number of times since boot that the RCU writer task
223 has changed the structure visible to readers.
225 o "tfle": If non-zero, indicates that the "torture freelist"
226 containing structures to be placed into the "rtc" area is empty.
227 This condition is important, since it can fool you into thinking
228 that RCU is working when it is not. :-/
230 o "rta": Number of structures allocated from the torture freelist.
232 o "rtaf": Number of allocations from the torture freelist that have
233 failed due to the list being empty. It is not unusual for this
234 to be non-zero, but it is bad for it to be a large fraction of
235 the value indicated by "rta".
237 o "rtf": Number of frees into the torture freelist.
239 o "rtmbe": A non-zero value indicates that rcutorture believes that
240 rcu_assign_pointer() and rcu_dereference() are not working
241 correctly. This value should be zero.
243 o "rtbe": A non-zero value indicates that one of the rcu_barrier()
244 family of functions is not working correctly.
246 o "rtbke": rcutorture was unable to create the real-time kthreads
247 used to force RCU priority inversion. This value should be zero.
249 o "rtbre": Although rcutorture successfully created the kthreads
250 used to force RCU priority inversion, it was unable to set them
251 to the real-time priority level of 1. This value should be zero.
253 o "rtbf": The number of times that RCU priority boosting failed
254 to resolve RCU priority inversion.
256 o "rtb": The number of times that rcutorture attempted to force
257 an RCU priority inversion condition. If you are testing RCU
258 priority boosting via the "test_boost" module parameter, this
259 value should be non-zero.
261 o "nt": The number of times rcutorture ran RCU read-side code from
262 within a timer handler. This value should be non-zero only
263 if you specified the "irqreader" module parameter.
265 o "Reader Pipe": Histogram of "ages" of structures seen by readers.
266 If any entries past the first two are non-zero, RCU is broken.
267 And rcutorture prints the error flag string "!!!" to make sure
268 you notice. The age of a newly allocated structure is zero,
269 it becomes one when removed from reader visibility, and is
270 incremented once per grace period subsequently -- and is freed
271 after passing through (RCU_TORTURE_PIPE_LEN-2) grace periods.
273 The output displayed above was taken from a correctly working
274 RCU. If you want to see what it looks like when broken, break
277 o "Reader Batch": Another histogram of "ages" of structures seen
278 by readers, but in terms of counter flips (or batches) rather
279 than in terms of grace periods. The legal number of non-zero
280 entries is again two. The reason for this separate view is that
281 it is sometimes easier to get the third entry to show up in the
282 "Reader Batch" list than in the "Reader Pipe" list.
284 o "Free-Block Circulation": Shows the number of torture structures
285 that have reached a given point in the pipeline. The first element
286 should closely correspond to the number of structures allocated,
287 the second to the number that have been removed from reader view,
288 and all but the last remaining to the corresponding number of
289 passes through a grace period. The last entry should be zero,
290 as it is only incremented if a torture structure's counter
291 somehow gets incremented farther than it should.
293 Different implementations of RCU can provide implementation-specific
294 additional information. For example, SRCU provides the following
297 srcu-torture: per-CPU(idx=1): 0(0,1) 1(0,1) 2(0,0) 3(0,1)
299 This line shows the per-CPU counter state. The numbers in parentheses are
300 the values of the "old" and "current" counters for the corresponding CPU.
301 The "idx" value maps the "old" and "current" values to the underlying
302 array, and is useful for debugging.
307 The following script may be used to torture RCU:
314 dmesg | grep torture:
316 The output can be manually inspected for the error flag of "!!!".
317 One could of course create a more elaborate script that automatically
318 checked for such errors. The "rmmod" command forces a "SUCCESS",
319 "FAILURE", or "RCU_HOTPLUG" indication to be printk()ed. The first
320 two are self-explanatory, while the last indicates that while there
321 were no RCU failures, CPU-hotplug problems were detected.