1RCU Torture Test Operation
   6The CONFIG_RCU_TORTURE_TEST config option is available for all RCU
   7implementations.  It creates an rcutorture kernel module that can
   8be loaded to run a torture test.  The test periodically outputs
   9status messages via printk(), which can be examined via the dmesg
  10command (perhaps grepping for "torture").  The test is started
  11when the module is loaded, and stops when the module is unloaded.
  15It is also possible to specify CONFIG_RCU_TORTURE_TEST=y, which will
  16result in the tests being loaded into the base kernel.  In this case,
  17the CONFIG_RCU_TORTURE_TEST_RUNNABLE config option is used to specify
  18whether the RCU torture tests are to be started immediately during
  19boot or whether the /proc/sys/kernel/rcutorture_runnable file is used
  20to enable them.  This /proc file can be used to repeatedly pause and
  21restart the tests, regardless of the initial state specified by the
  24You 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
  26this can sometimes be useful in finding boot-time bugs.
  31This module has the following parameters:
  33fqs_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.
  39fqs_holdoff     Holdoff time (in microseconds) between consecutive calls
  40                to force_quiescent_state() within a burst.
  42fqs_stutter     Wait time (in seconds) between consecutive bursts
  43                of calls to force_quiescent_state().
  45irqreader       Says to invoke RCU readers from irq level.  This is currently
  46                done via timers.  Defaults to "1" for variants of RCU that
  47                permit this.  (Or, more accurately, variants of RCU that do
  48                -not- permit this know to ignore this variable.)
  50n_barrier_cbs   If this is nonzero, RCU barrier testing will be conducted,
  51                in which case n_barrier_cbs specifies the number of
  52                RCU callbacks (and corresponding kthreads) to use for
  53                this testing.  The value cannot be negative.  If you
  54                specify this to be non-zero when torture_type indicates a
  55                synchronous RCU implementation (one for which a member of
  56                the synchronize_rcu() rather than the call_rcu() family is
  57                used -- see the documentation for torture_type below), an
  58                error will be reported and no testing will be carried out.
  60nfakewriters    This is the number of RCU fake writer threads to run.  Fake
  61                writer threads repeatedly use the synchronous "wait for
  62                current readers" function of the interface selected by
  63                torture_type, with a delay between calls to allow for various
  64                different numbers of writers running in parallel.
  65                nfakewriters defaults to 4, which provides enough parallelism
  66                to trigger special cases caused by multiple writers, such as
  67                the synchronize_srcu() early return optimization.
  69nreaders        This is the number of RCU reading threads supported.
  70                The default is twice the number of CPUs.  Why twice?
  71                To properly exercise RCU implementations with preemptible
  72                read-side critical sections.
  75                The number of seconds between each attempt to execute a
  76                randomly selected CPU-hotplug operation.  Defaults to
  77                zero, which disables CPU hotplugging.  In HOTPLUG_CPU=n
  78                kernels, rcutorture will silently refuse to do any
  79                CPU-hotplug operations regardless of what value is
  80                specified for onoff_interval.
  82onoff_holdoff   The number of seconds to wait until starting CPU-hotplug
  83                operations.  This would normally only be used when
  84                rcutorture was built into the kernel and started
  85                automatically at boot time, in which case it is useful
  86                in order to avoid confusing boot-time code with CPUs
  87                coming and going.
  90                The number of seconds to keep the test threads affinitied
  91                to a particular subset of the CPUs, defaults to 3 seconds.
  92                Used in conjunction with test_no_idle_hz.
  94shutdown_secs   The number of seconds to run the test before terminating
  95                the test and powering off the system.  The default is
  96                zero, which disables test termination and system shutdown.
  97                This capability is useful for automated testing.
  99stall_cpu       The number of seconds that a CPU should be stalled while
 100                within both an rcu_read_lock() and a preempt_disable().
 101                This stall happens only once per rcutorture run.
 102                If you need multiple stalls, use modprobe and rmmod to
 103                repeatedly run rcutorture.  The default for stall_cpu
 104                is zero, which prevents rcutorture from stalling a CPU.
 106                Note that attempts to rmmod rcutorture while the stall
 107                is ongoing will hang, so be careful what value you
 108                choose for this module parameter!  In addition, too-large
 109                values for stall_cpu might well induce failures and
 110                warnings in other parts of the kernel.  You have been
 111                warned!
 114                The number of seconds to wait after rcutorture starts
 115                before stalling a CPU.  Defaults to 10 seconds.
 117stat_interval   The number of seconds between output of torture
 118                statistics (via printk()).  Regardless of the interval,
 119                statistics are printed when the module is unloaded.
 120                Setting the interval to zero causes the statistics to
 121                be printed -only- when the module is unloaded, and this
 122                is the default.
 124stutter         The length of time to run the test before pausing for this
 125                same period of time.  Defaults to "stutter=5", so as
 126                to run and pause for (roughly) five-second intervals.
 127                Specifying "stutter=0" causes the test to run continuously
 128                without pausing, which is the old default behavior.
 130test_boost      Whether or not to test the ability of RCU to do priority
 131                boosting.  Defaults to "test_boost=1", which performs
 132                RCU priority-inversion testing only if the selected
 133                RCU implementation supports priority boosting.  Specifying
 134                "test_boost=0" never performs RCU priority-inversion
 135                testing.  Specifying "test_boost=2" performs RCU
 136                priority-inversion testing even if the selected RCU
 137                implementation does not support RCU priority boosting,
 138                which can be used to test rcutorture's ability to
 139                carry out RCU priority-inversion testing.
 142                The number of seconds in an RCU priority-inversion test
 143                cycle.  Defaults to "test_boost_interval=7".  It is
 144                usually wise for this value to be relatively prime to
 145                the value selected for "stutter".
 148                The number of seconds to do RCU priority-inversion testing
 149                within any given "test_boost_interval".  Defaults to
 150                "test_boost_duration=4".
 152test_no_idle_hz Whether or not to test the ability of RCU to operate in
 153                a kernel that disables the scheduling-clock interrupt to
 154                idle CPUs.  Boolean parameter, "1" to test, "0" otherwise.
 155                Defaults to omitting this test.
 157torture_type    The type of RCU to test, with string values as follows:
 159                "rcu":  rcu_read_lock(), rcu_read_unlock() and call_rcu().
 161                "rcu_sync":  rcu_read_lock(), rcu_read_unlock(), and
 162                        synchronize_rcu().
 164                "rcu_expedited": rcu_read_lock(), rcu_read_unlock(), and
 165                        synchronize_rcu_expedited().
 167                "rcu_bh": rcu_read_lock_bh(), rcu_read_unlock_bh(), and
 168                        call_rcu_bh().
 170                "rcu_bh_sync": rcu_read_lock_bh(), rcu_read_unlock_bh(),
 171                        and synchronize_rcu_bh().
 173                "rcu_bh_expedited": rcu_read_lock_bh(), rcu_read_unlock_bh(),
 174                        and synchronize_rcu_bh_expedited().
 176                "srcu": srcu_read_lock(), srcu_read_unlock() and
 177                        call_srcu().
 179                "srcu_sync": srcu_read_lock(), srcu_read_unlock() and
 180                        synchronize_srcu().
 182                "srcu_expedited": srcu_read_lock(), srcu_read_unlock() and
 183                        synchronize_srcu_expedited().
 185                "srcu_raw": srcu_read_lock_raw(), srcu_read_unlock_raw(),
 186                        and call_srcu().
 188                "srcu_raw_sync": srcu_read_lock_raw(), srcu_read_unlock_raw(),
 189                        and synchronize_srcu().
 191                "sched": preempt_disable(), preempt_enable(), and
 192                        call_rcu_sched().
 194                "sched_sync": preempt_disable(), preempt_enable(), and
 195                        synchronize_sched().
 197                "sched_expedited": preempt_disable(), preempt_enable(), and
 198                        synchronize_sched_expedited().
 200                Defaults to "rcu".
 202verbose         Enable debug printk()s.  Default is disabled.
 207The statistics output is as follows:
 209        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
 210        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
 211        rcu-torture: Reader Pipe:  727860534 34213 0 0 0 0 0 0 0 0 0
 212        rcu-torture: Reader Batch:  727877838 17003 0 0 0 0 0 0 0 0 0
 213        rcu-torture: Free-Block Circulation:  155440 155440 155440 155440 155440 155440 155440 155440 155440 155440 0
 214        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
 216The command "dmesg | grep torture:" will extract this information on
 217most systems.  On more esoteric configurations, it may be necessary to
 218use other commands to access the output of the printk()s used by
 219the RCU torture test.  The printk()s use KERN_ALERT, so they should
 220be evident.  ;-)
 222The first and last lines show the rcutorture module parameters, and the
 223last line shows either "SUCCESS" or "FAILURE", based on rcutorture's
 224automatic determination as to whether RCU operated correctly.
 226The entries are as follows:
 228o       "rtc": The hexadecimal address of the structure currently visible
 229        to readers.
 231o       "ver": The number of times since boot that the RCU writer task
 232        has changed the structure visible to readers.
 234o       "tfle": If non-zero, indicates that the "torture freelist"
 235        containing structures to be placed into the "rtc" area is empty.
 236        This condition is important, since it can fool you into thinking
 237        that RCU is working when it is not.  :-/
 239o       "rta": Number of structures allocated from the torture freelist.
 241o       "rtaf": Number of allocations from the torture freelist that have
 242        failed due to the list being empty.  It is not unusual for this
 243        to be non-zero, but it is bad for it to be a large fraction of
 244        the value indicated by "rta".
 246o       "rtf": Number of frees into the torture freelist.
 248o       "rtmbe": A non-zero value indicates that rcutorture believes that
 249        rcu_assign_pointer() and rcu_dereference() are not working
 250        correctly.  This value should be zero.
 252o       "rtbe": A non-zero value indicates that one of the rcu_barrier()
 253        family of functions is not working correctly.
 255o       "rtbke": rcutorture was unable to create the real-time kthreads
 256        used to force RCU priority inversion.  This value should be zero.
 258o       "rtbre": Although rcutorture successfully created the kthreads
 259        used to force RCU priority inversion, it was unable to set them
 260        to the real-time priority level of 1.  This value should be zero.
 262o       "rtbf": The number of times that RCU priority boosting failed
 263        to resolve RCU priority inversion.
 265o       "rtb": The number of times that rcutorture attempted to force
 266        an RCU priority inversion condition.  If you are testing RCU
 267        priority boosting via the "test_boost" module parameter, this
 268        value should be non-zero.
 270o       "nt": The number of times rcutorture ran RCU read-side code from
 271        within a timer handler.  This value should be non-zero only
 272        if you specified the "irqreader" module parameter.
 274o       "Reader Pipe": Histogram of "ages" of structures seen by readers.
 275        If any entries past the first two are non-zero, RCU is broken.
 276        And rcutorture prints the error flag string "!!!" to make sure
 277        you notice.  The age of a newly allocated structure is zero,
 278        it becomes one when removed from reader visibility, and is
 279        incremented once per grace period subsequently -- and is freed
 280        after passing through (RCU_TORTURE_PIPE_LEN-2) grace periods.
 282        The output displayed above was taken from a correctly working
 283        RCU.  If you want to see what it looks like when broken, break
 284        it yourself.  ;-)
 286o       "Reader Batch": Another histogram of "ages" of structures seen
 287        by readers, but in terms of counter flips (or batches) rather
 288        than in terms of grace periods.  The legal number of non-zero
 289        entries is again two.  The reason for this separate view is that
 290        it is sometimes easier to get the third entry to show up in the
 291        "Reader Batch" list than in the "Reader Pipe" list.
 293o       "Free-Block Circulation": Shows the number of torture structures
 294        that have reached a given point in the pipeline.  The first element
 295        should closely correspond to the number of structures allocated,
 296        the second to the number that have been removed from reader view,
 297        and all but the last remaining to the corresponding number of
 298        passes through a grace period.  The last entry should be zero,
 299        as it is only incremented if a torture structure's counter
 300        somehow gets incremented farther than it should.
 302Different implementations of RCU can provide implementation-specific
 303additional information.  For example, SRCU provides the following
 304additional line:
 306        srcu-torture: per-CPU(idx=1): 0(0,1) 1(0,1) 2(0,0) 3(0,1)
 308This line shows the per-CPU counter state.  The numbers in parentheses are
 309the values of the "old" and "current" counters for the corresponding CPU.
 310The "idx" value maps the "old" and "current" values to the underlying
 311array, and is useful for debugging.
 316The following script may be used to torture RCU:
 318        #!/bin/sh
 320        modprobe rcutorture
 321        sleep 3600
 322        rmmod rcutorture
 323        dmesg | grep torture:
 325The output can be manually inspected for the error flag of "!!!".
 326One could of course create a more elaborate script that automatically
 327checked for such errors.  The "rmmod" command forces a "SUCCESS",
 328"FAILURE", or "RCU_HOTPLUG" indication to be printk()ed.  The first
 329two are self-explanatory, while the last indicates that while there
 330were no RCU failures, CPU-hotplug problems were detected.
 331 kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.